Multi-label classification with local pairwise and high-order label correlations using graph partitioning

In multi-label learning problems, the class labels are correlated and label correlations can be leveraged to improve predictive performance of a classifier. Methods that consider high-order in space, usually do not utilize pairwise correlations. most these methods, considered as prior knowledge, which misleading problems with noisy or missing labels. such cases, correlation part model training task is more effective. this paper, rule-based evolutionary classification method proposed incorporates local through subsets dependencies. Graph structures employed dependencies estimated similarities used obtain accurate sets for rules. To refine relations, novel hierarchical density-based clustering k-way partitioning graphs based on their The effectiveness experimented multiple benchmark datasets from different domains compared several well-known algorithms. has shown highest average rank along metrics results consistently better than similar methods statistical significance.